US 20080077014 A1
A method and system for utilizing shape analysis to assess fetal abnormality. According to one embodiment, coordinates of points identifying a shape in a fetal image are received, coefficients of one or more mathematical functions that describe the identified shape are determined, and the determined coefficients are utilized as markers to assess fetal abnormality.
1. A computer-implemented method for utilizing shape analysis to assess human fetal abnormality, comprising:
receiving coordinates of points identifying an anatomical shape in a human fetal image;
determining a plurality of coefficients of one or more mathematical functions that describe the identified anatomic shape; and
utilizing the determined plurality of coefficients as markers to assess human fetal abnormality.
2. A method of
3. A method of
4. A method of
5. A method of
6. A method of
7. A method of
8. A method of
9. A method of
10. A method of
11. A method of
12. A method of
13. A method of
14. A method of
15. A method of
16. A method of
17. A method of
18. A method of
19. A method of
20. A method of
adjusting the received coordinates to align the shape according to a particular axis before the coefficients are determined.
21. A method of
adjusting the received coordinates before the coefficients are determined by at least one of translating the coordinates, rotating the coordinates and scaling the coordinates.
22. A method of
23. A method of
24. An apparatus for utilizing shape analysis to assess human fetal abnormality, comprising:
a processor; and
a memory storing instructions adapted to be executed by said processor to:
receive coordinates of points identifying an anatomical shape in a human fetal image;
determine a plurality of coefficients of one or more mathematical functions that describe the identified anatomic shape; and
utilize the determined coefficients as markers to assess human fetal abnormality.
25. An apparatus according to
26. An apparatus according to
27. An apparatus according to
28. An apparatus according to
29. An apparatus according to
30. A system for utilizing shape analysis to assess human fetal abnormality, comprising:
a means for receiving coordinates of points identifying an anatomic shape in a human fetal image;
a means for determining a plurality of coefficients of one or more mathematical functions that describe the identified anatomic shape; and
a means for utilizing the determined coefficients as markers to assess human fetal abnormality.
31. A system according to
32. A system according to
33. A system according to
34. A system according to
35. A system according to
This application claims the benefit under 35 U.S.C. § 119(e) of U.S. Provisional Application No. 60/490,540, filed Jul. 29, 2003 and U.S. Provisional Application No. 60/493,442, filed Aug. 8, 2003, both of which are hereby incorporated by reference as if repeated herein in their entirety.
Prenatal screening methods are routinely employed to assess the likelihood of fetal abnormalities, commonly referred to as birth defects. For example, Down syndrome or Trisomy 21 is the most common cause of severe learning disability and accounts for approximately one half of all chromosomal anomalies in live born children.
Current methods to screen prenatally for trisomy 21 involve maternal serum testing for biochemical markers and/or ultrasound evaluation of biophysical markers. Maternal serum screening involves the quantitative analysis of biochemical markers and risk assessment based on likelihood ratios derived from the population distributions of affected and unaffected pregnancies. Ultrasound evaluation, however, has historically involved visual observation of a fetal image and deciding empirically whether the image looks “normal” or “abnormal” (for example, whether the cerebellum appears as a banana sign for open spina bifida). This approach requires extensive experience in the “art” of ultrasound and the interpretation is necessarily subjective.
Accordingly, there is a need in the art for a system and method that adequately evaluates the morphological changes observed with birth defects during prenatal screening.
Embodiments of the present invention provide for utilizing shape analysis to assess fetal abnormality. According to one embodiment, coordinates of points identifying a shape in a fetal image are received, coefficients of one or more mathematical functions that describe the identified shape are determined, and the determined coefficients are utilized as markers to assess fetal abnormality.
As shown in
The use of multidimensional coordinates allows for the evaluation of a shape as a whole. In one embodiment, a statistical shape analysis involves the tracing of an outline around the part of a fetal image to be analyzed. The points that make up this curve are then analyzed to derive a function that best fits the individualized points around the outline. The coefficients of this function may be considered random variables and can be determined for each evaluated image. The coefficients may then be analyzed using multivariate statistics to determine if they are outliers compared to the normal population. A subject shape that has coefficients outside the normal ranges observed in control shapes would indicate that the subject shape was significantly different than expected.
Described below are several embodiments within which the present invention may be implemented.
Input device 320 may include a keyboard, mouse, pen-operated touch screen or monitor, voice-recognition device, or any other device that accepts input. Output device 330 may include a monitor, printer, disk drive, speakers, or any other device that provides output.
Storage device 340 may include volatile and nonvolatile data storage, including one or more electrical, magnetic or optical memories such as a RAM, cache, hard drive, CD-ROM drive, tape drive or removable storage disk. Communication device 360 may include a modem, network interface card, or any other device capable of transmitting and receiving signals over a network. The components of user computing device 300 may be connected via an electrical bus or wirelessly.
Client software 350 may be stored in storage device 340 and executed by processor 310, and may include, for example, imaging and analysis software that embodies the functionality of the present invention.
Network link 415 may include telephone lines, DSL, cable networks, T1 or T3 lines, wireless network connections, or any other arrangement that implements the transmission and reception of network signals. Network 410 may include any type of interconnected communication system, and may implement any communications protocol, which may secured by any security protocol.
Server 420 includes a processor and memory for executing program instructions, as well as a network interface, and may include a collection of servers. In one particular embodiment, server 420 may include a combination of servers such as an application server and a database server. Database 440 may represent a relational or object database, and may be accessed via server 420.
User computing device 300 and server 420 may implement any operating system, such as Windows or UNIX. Client software 350 and server software 430 may be written in any programming language, such as ABAP, C, C++, Java or Visual Basic.
According to this embodiment, a transverse view of the fetal head is obtained by ultrasound (e.g., UCD 300 a), saved as a bit-map image and transferred to another computer (e.g., UCD 300 b). Then, in step 500, a user (e.g., user 400 b) employs digitizing software (e.g., client software 350), such as TPSDIG, DigitX, CalExcel; DSDigit, Digical, Windig or MacMorph, to create an outline of the fetal head to be analyzed. (Digitizing software provides coordinate data when a user clicks on a particular point in a bit-map image.) In creating the outline, the user may identify two landmarks on the bit-map image along the OFD (ocipito-frontal diameter) axis so that the image may be aligned against a consistent axis and to allow for a uniform assessment of points on each image.
As shown in
In step 510, the user-identified outline points are converted into data coordinates by the digitizing software. TABLE 1 represents the landmarks and outline points of
In order to ensure the coordinate data is aligned properly with the two specified landmarks, the user may employ a software program that utilizes general algorithms to adjust the coordinate data by rotating and translating each data point so that the OFD axis lies along the horizontal axis and that the first point lies at the origin, (x,y)=(0,0). These adjustments do not change the shape of the outline being investigated. The software program may employ the following general algorithm:
The set of FinalX, FinalY values represent the rotated data points such that the OFD axis lies horizontally. In different scenarios, it may be appropriate to rotate the image in a counter clockwise manner; if this is the case then the formulas for FinalX and FinalY are:
In step 520, the user may employ software such as the NTSYSPC program (Exeter Software), which uses an elliptical Fourier analysis program to generate a series of best fit curves to the coordinate data using harmonics. The zero harmonic consists of only 2 coefficients and represents only translation (i.e. movement in the x and y direction) and not shape itself so these 2 coefficients are usually not analyzed. Each harmonic comprises four coefficients of interpolation functions that describe the user-identified outline shape. In general, the more harmonics used, the better the fit to the coordinates; however, if too many harmonics are used the fit may be too good since small differences might appear due to poor placement of the outline point during step 500. In this embodiment, the user evaluates the first harmonic (i.e., the four coefficients a1, b1, c1 and d1).
The user may choose alternatives to the method described above for aligning the coordinate data. For example, the user could rotate the image so that the long axis of the shape described by the first harmonic is parallel to the x-axis, or the user can align the image such that the first point of the outline is set equal to the end point of the first harmonic to the x-axis, or the user could combine both of these methods. These options are available as part of the Elliptical Fourier Analysis module in the NTSYSPC program.
In step 530, the determined coefficients may be utilized as markers to assess fetal abnormality by conducting a statistical analysis to compare the patient's determined coefficients with reference parameters derived from reference data of coefficient markers (e.g., a statistical distribution of determined coefficients in the unaffected population and/or affected population). One exemplary method of doing this is to calculate the Mahalanobis Squared Distance (MSD) value for the patient's coefficients.
Prior to determining the outline coefficients, the coordinates may be adjusted by scaling the coordinates so that the area enclosed inside the outline equals 1. The original areas of the enclosed outline could be included as a separate variable in addition to the coefficients as part of the statistical comparison analysis. Also, many shapes change with growth in the fetus. As a result, it may be necessary to adjust the observed coefficients to account for gestational age of the fetus as part of the statistical analysis.
Several other shape analysis methods besides elliptical Fourier analysis may be utilized by the present invention. All of the following methods determine coefficients of functions that may be used as markers to assess fetal abnormality. Some examples of outline methods are:
B. cubic splines
C. parametric polynomials
D. parametric cubic splines
E. bezier curves
F. Fourier analysis of equally spaced radii
G. dual axis Fourier analysis
All of these methods deal with outlines, of which there are four classes—simple open outlines, simple closed outlines, complex open outlines, complex closed outlines. A simple open outline is an open outline that has only one value of y for each x. A complex open outline is an outline that can have more than one y value for each x. Simple closed outlines are closed outlines such that if one draws a line from the center through the outline it only crosses the outline once whereas in a complex outline the line would cross the outline more than once. In most cases an open outline can be analyzed as if it is a closed outline by assuming there is a straight line from the last to first point of the outline or by mirroring all of the coordinates around the x (or y) axis, however the outline can be analyzed as an open outline. Some of the methods above, such as the polynomial method, work with open outlines while others like the elliptical Fourier analysis work with closed outlines. It is also possible to analyze 3-dimensional outlines (x,y,z) in accordance with an alternative embodiment of the present invention.
To provide the reference data to which the determined coefficients are compared in step 530, a statistical algorithm may be utilized to determine the statistical distribution of the coefficients in the unaffected population using a set of coefficients for a series of unaffected pregnancies. The distribution may be defined by a series of reference parameters for each coefficient or pair of coefficients such as means, standard deviations and correlations. The coefficients associated with an outline in a particular patient could be compared to these reference parameters and the chance that the coefficients could be equal to or more extreme than their observed values could be determined.
If affected cases are available then they may be included in the reference data set to determine the statistical distribution of the coefficients in the affected population using a set of coefficients from a series of affected pregnancies. The patient's coefficients could then be compared to the distribution of coefficients in the affected population. Alternatively, the patient's coefficients could be compared to both the unaffected distribution and the affected distribution as defined by the reference parameters. For example, if both the unaffected cases and affected cases are multivariately normally distributed, the distribution function for the multivariate normal distribution can be used with the reference parameters for the unaffected distribution to determine a relative frequency for the unaffected distribution and then again for the affected distribution to determine a relative frequency for the affected distribution, and then a likelihood ratio may be determined. A likelihood ratio equals the quotient of the relative frequency in the affected distribution to the relative frequency in the unaffected distribution. A risk result (e.g., 1 in 100) gives the chance that a patient with the same parameters could have a child with a fetal abnormality. A likelihood ratio gives the relative risk that the patient could have a child with a fetal abnormality.
The likelihood ratio can be used to multiply a prior risk to determine a posterior risk (after accounting for the minor adjustment between odds and risk, if necessary). For example, the prior risk of Down syndrome is often based on maternal age. If the statistical distribution of the determined coefficients are independent of the distribution of other markers, the likelihood ratio could also be used to adjust the risk of Down syndrome determined by the other markers to determine the overall risk of Down syndrome based on the outline and the other markers. Alternatively, if the coefficients are not independent of the other markers, then risk of Down syndrome could be determined by utilizing reference parameters for a combination of the coefficients and the other markers together using multivariate normal distributions or other distribution functions.
Examples of other markers include nuchal translucency, free Beta hCG and PAPP-A, Ductus Venosus, absent or hypoplastic nasal bone observed on ultrasound, maternal blood alpha-fetoprotein, maternal blood hCG, maternal blood unconjugated estriol, maternal blood dimeric inhibin A, maternal urine total estriol, maternal urine beta core fragment, maternal urine hyperglycosylated hCG, maternal blood hyperglycosylated hCG, ultrasound “soft markers” which include for example, nuchal edema or increased nuchal fold, short femur, hyperechogenic bowel, and echogenic foci in the heart, etc.
As mentioned above, the other markers can be combined statistically with the results from the shape analysis to provide a final result to the patient. Alternatively, the medical tests for the other markers could be performed prior to the ultrasound exam and then, when the ultrasound exam is completed, the results of the other marker tests can be thereafter combined with the results from the shape analysis.
For providing reference data according to this embodiment of the present invention, columns 2-5 of TABLE 2A show the results of evaluating the first harmonic (i.e., the four coefficients a1, b1, c1 and d1) of fetal head outlines in a study of 35 unaffected pregnancies in the first trimester. TABLE 2B lists the reference parameters consisting of the mean and standard deviation of each coefficient and the variance/covariance matrix consisting of the variance (standard deviation squared) and covariance between each pair of coefficients and the formula for calculating a Mahalanobis-Squared Distance (MSD) for each case.
TABLE 2C shows the MSD calculation in the 35 unaffected cases. A cut-off beyond 95% of the observed data was established representing a point halfway between the last 2 MSD values (11.714). TABLE 2D shows the results of 2 patients who happened to be carrying a fetus with Down syndrome, both of which based on their MSD calculation are outliers. As more data from Down syndrome pregnancies are gathered, additional reference parameters (e.g., means, standard deviations, and covariances) based on the Down syndrome cases could be calculated along with other statistical techniques such as likelihood ratios to determine the odds that a patient is carrying a fetus with Down syndrome.
In this embodiment of the present invention, only the data associated with TABLE 2B would have to be stored in the computing device for the statistical comparison analysis of a particular patient to be conducted. This would preserve storage resources in the event of reference data based on very large populations.
3D ultrasound allows for the simultaneous visualization of the fetus in 3 separate 2-D planes, as shown in
Misalignment of an image in any given plane distorts the images in the other two planes. Therefore, in order to perform proper 2-D image analysis of any given plane, the image should first be aligned properly in all three planes. For example, to analyze an image in the sagittal plane, a user can assure a proper view by aligning the fetus in the coronal and transverse views. In addition, by choosing a landmark in each plane (e.g. the white dot placed at the bridge of the nose in the
The use of 3D sonography is a recent advance in prenatal ultrasound. The technique generates a multiplanar display of separate images in the coronal, sagittal and transverse planes obtained by the ascertainment of a single “volume”. Once the volume is obtained the images in each plane may be rotated to provide consistent and reproducible planes as part of a “post processing” evaluation. This advance was previously unobtainable using conventional 2D sonographic techniques. Once a given desired image and plane is obtained, it may be superimposed on a digitized screen where geometric morphometric analysis may be performed.
According to this embodiment, a fetal brow is analyzed in the sagittal plane—from the bridge of the nose to the midportion of the top of the skull. The fetal image of
In step 800, a user (e.g., user 400 b) employs digitizing software to create an outline of the fetal brow to be analyzed. As illustrated in
The first line in
In step 810, the user-identified outline points are converted into data coordinates by the digitizing software. The user may choose to use only the first eight points in subsequent analysis since these points more determinatively represent the fetal brow.
In step 820, an elliptical Fourier analysis is employed using three harmonics (i.e., three sets of coefficients). The determined coefficients are utilized as markers to assess fetal abnormality by conducting a statistical analysis, such as a principal component (PC) analysis.
In step 830, PC scores are determined based on a PC analysis of the determined Fourier coefficients, and in step 840, the MSD values of the PC scores are calculated. As discussed above, this shape analysis may also be combined with other markers to more completely assess fetal abnormality.
The PC analysis utilized by this embodiment is a statistical technique used with multivariate data to reduce the number of variables used in further statistical analysis. (It can also be used as an exploratory analysis to see which of the variables are most important). The PC analysis is a standard statistical technique that generates a set of linear combinations of the underlying variables. These linear combinations represent new variables that can be used in other statistical analyses. The first linear combination is the most important variable and so on. Higher numbered principal components can be dropped from further analyses since they tend to represent noise. Once the principle components are determined (e.g., 0.3X1+0.2X2+0.4X3 . . . ), the PC scores can then be calculated which can be used as the variables in an MSD calculation.
For providing reference data according to this embodiment of the present invention, TABLE 4 shows the results of evaluating the first three harmonics (i.e., three sets of the four coefficients A, B, C and D) of fetal brow outlines in a study of 32 normal pregnancies.
TABLE 5 shows the resulting PC analysis of the Fourier coefficients.
In this embodiment, the first 3 principal components are retained and used to calculate the associated PC variables. First a z-score for each coefficient is calculated by subtracting the mean of the coefficient and then dividing the difference by the standard deviation of the coefficient from TABLE 5. Each PC variable is then created by multiplying the scoring coefficient for each (a1-d3) by the z-score for each variable and then summing the 12 products, as shown in TABLE 6.
The PC variables are used in an MSD calculation to create an atypicality index, as shown in TABLE 7. The PC variables are independent of each other so the variance/covariance matrix is not needed. A cut-off of MSD>7.81 (95th percentile of the expected X2 with 3 d.o.f.) was used to define outliers.
TABLE 8A shows the coefficients from an elliptical Fourier analysis for 6 affected cases and 8 images that represent an additional image from a patient in the reference set. The MSD is calculated using the means and standard deviations of the coefficients, the PC scoring coefficients and the MSD formula for the 32 unaffected patients, the six affected patients and 8 additional images from patients in the group. The results are shown in TABLE 8B. The 3 outliers were all abnormal (trisomy 21, trisomy 18, and one fetus with multiple congenital anomalies). Two trisomy 21 and one case with an unbalanced translocation were not outliers. The affected cases were 1-3× Trisomy 18, 2-21×2 campomelic dysplasia, 2-23×2, 2-23×3, 2-23×4 (same patient) Trisomy 21, 865×2 Trisomy 21, 865-5×2 Trisomy 21, 865-17×2 translocation.
In this embodiment of the present invention, only the scoring coefficients, means and standard deviations of the coefficients from TABLE 5 and the standard deviations of the PC variables in TABLE 7 would have to be stored in the computing device for the statistical comparison analysis of a particular patient to be conducted. This would preserve storage resources in the event of reference data based on very large populations.
Several embodiments of the invention are specifically illustrated and/or described herein. However, it will be appreciated that modifications and variations of the invention are covered by the above teachings and within the purview of the appended claims without departing from the spirit and intended scope of the invention.